1. Technical Field
The present invention relates generally to computer data network equipment and methods, and more particularly to balancing the loading amongst distributed network servers by controlling the conversion of domain names to IP-addresses in domain name server equipment. The invention selects the load distribution criteria based on a unique algorithm.
2. Description of the Prior Art
The world wide web (WWW), and especially the Internet, are quickly becoming the principle way businesses sell products and communicate with customers and suppliers. Some now call the Internet a xe2x80x9cmission-critical business delivery infrastructure.xe2x80x9d As a consequence, intemet servers and so-called xe2x80x9cIntranetxe2x80x9d servers are worked harder than ever before. The number of clients servers now must support has increased dramatically. Intranet servers must now be able to service hundreds of simultaneous client requests, while their extemal-counterpart Internet servers must be able to support tens of thousands of simultaneous client connections.
Clients demand and expect rapid response and a 7-day and week, 24-hours a day (xe2x80x9c7xc3x9724xe2x80x9d) availability. Mission-critical web-computing infrastructures must be able to dynamically scale server capacity to match aggregate client demand and still ensure continuous service availability. One way to do just that has been to run each application on several servers, and then continually balance the client loading on the various servers, e.g., xe2x80x9cserver load balancing.xe2x80x9d
Server load balancers use information in the Layer 3 and Layer 4 packet headers to identify and manage application-layer sessions. For example, TCP or UDP port numbers, the SYN/FIN bits that mark the start and end of TCP application sessions and IP source and destination addresses.
Traditional server load balancers are PC-based software products with limited performance and connectivity. The rapid growth in traffic volume and server population is giving rise to a new generation of switch-integrated server load balancers that offer many orders of magnitude improvements in performance, connectivity, resiliency and economy.
A new generation of switch-based server load balancers consolidates multiple web infrastructure functions and load balancing application servers with multi-layer switching, e.g., redirection traffic to caches, load balancing traffic to multiple firewalls, packet filtering and bandwidth management.
Alteon WebSystems coined the term xe2x80x9cServer Switchxe2x80x9d to represent this new class of device that front-ends server farms and provide server-related traffic management in all mission critical intemet/Intranet infrastructures. Server Switches dynamically distribute application load across a group of servers running a common application (or set of applications) while making the group appear as one server to the network. A number of web servers with access to the same content can be logically combined into an HTTP hunt group, which is a group of servers that supports a common application or set of applications. The hunt group provides a xe2x80x9cvirtualxe2x80x9d HTTP service to clients. Clients are not aware that there are a number of real servers participating in providing this service. The clients access the service using a virtual service address that resides in a server switch that front-ends the real servers. As connection requests arrive for the virtual service, the server switch passes these requests onto one of the real servers in the hunt group based upon knowledge of the servers"" availability, load handling capability, and present load.
In this way, multiple servers can be used to achieve the total amount of application processing capacity demanded by the users of the system. Each new server adds its capacity to the pool of processing power available for the application.
Equally important, as servers go out of service due either to failure or maintenance operations, the remaining healthy servers pick up the load with little or no perceived impact to users. To achieve this, the server switch must continuously monitor the health of all servers and each application to which it distributes client load. The server switches must also support hot-standby configurations for complete systems redundancy.
A key part of server load balancing is session management. Once a session request is assigned to a real server, the server switch must recognize all successive packets associated with that session. These packets are processed and forwarded appropriately to make sure that the client continues to be associated with the same physical server for the duration of each session.
Server switches also monitor the completion of sessions at which time the binding of the connection to the physical server can be removed. This ensures that the next time a client connects, he is preferably connected to the most available server at the time, providing the best possible service to each client. Special mechanisms can be invoked by the administrator if the application requires successive connections to be forwarded to the same physical server, such as with FTP control and data connections, SSL (Secure Sockets Layer), and persistent HTTP used for multi-page forms and search engines.
Environments that benefit from server load balancing include web hosting services, on-line service providers and corporate data centers with high availability requirements. In theory, server load balancing can be used to support any TCP-based or UDP-based application where common content is available across a group of servers. In practice, servers supporting Intemet/Intranet applications, such as web servers, FTP servers, domain name server servers and RADIUS servers is preferably the first to take advantage of server load balancing to support the high growth and unpredictable volume of web-oriented traffic.
The majority of web pages contain read-only information. This makes web-hosting environments ideal for server load balancing. Web hosts and on-line service providers typically deploy multiple HTTP, FTP and other application servers today, with load distributed across them statically, or more commonly, via round-robin domain name server. Both methods are undesirable because they are not fault-tolerant and require a high degree of administration. Server load balancing enables transparent use of multiple servers with built-in high availability support.
Many clustering systems today provide superior failover capabilities but offer no load-balancing support. Some systems also limit the number of servers that can participate in a cluster. These constraints impact the scalability of the clustering solutions. Server load balancing enables flexible coupling of servers into load-sharing hunt groups. It also improves server utilization efficiency by enabling redundant servers to share load.
More often than not, server environments today are multi-vendor and multi-OS. Popular clustering solutions today are limited to servers from a single vendor or servers running a single operating system. Server load balancing on a server switch enables heterogeneous servers supporting TCP and UDP applications to be loosely coupled in a load-sharing cluster, maximizing server investment returns.
An actual Internet web-site that serves the web-pages to a client in response to a URL domain name is automatically and transparently selected from a list of many distributed sites each having identical data storage. In a peer hand-off process, a switch receives domain name server lookup request for a particular domain name. The switch examines the source IP-address for the domain name server request, examines the user""s IP-address, and determines if there is server site that is geographically close to that user. The switch examines an ordered hand-off table corresponding to the domain. The switch chooses a next remote server (or one of its own virtual Internet protocol addresses) based on, (a) the remote server location compared to domain name server request source, (b) the remote servers"" weights, and (c) the remote server that experienced the previous hand-off. The switch then sends the domain name server response back to client domain name server with the IP-addresses in an ordered list.